机器学习知识积累

判别模型(discriminative) vs生成模型(generative)

逻辑回归算法简单，对特征工程的要求就非常高。必须做特征归一化，否则各特征重要程度不一。

http://www.cnblogs.com/maybe2030/p/6336896.html

特征正规化

from sklearn import preprocessing
df['normized'] = preprocessing.scale(df['non-normalized'])

L1 Distance/L1-Norm .vs. L2 Distance/L2-Norm

L1 = sum(abs(t1_i-t2_i))

L1-norm = |A|+|B_1|+... + |b_n|

L2 = sqrt(sum(t1_i-t2_i)^2)